1 A One Lecture Introduction to Fredholm Operators Craig Sinnamon 250380771 December, 2009 1. Introduction This is an elementary introdution to Fredholm operators on a Hilbert space H. Fredholm operators are named after a Swedish mathematician, Ivar Fredholm(1866-1927), who studied integral equations. We will introduce two definitions of a Fredholm operator and prove their equivalance. We will also discuss briefly the index map defined on the set of Fredholm operators. Note that results proved in 4154A Functional Analysis by Prof. Khalkhali may be used without proof. Definition 1.1: A bounded, linear operator T : H ! H is said to be Fredholm if 1) rangeT ⊆ H is closed 2) dimkerT < 1 3) dimketT ∗ < 1 Notice that for a bounded linear operator T on a Hilbert space H, kerT ∗ = (ImageT )?. Indeed: x 2 kerT ∗ , T ∗x = 0 ,< T ∗x; y >= 0 for all y 2 H ,< x:T y >= 0 , x 2 (ImageT )? A good way to think about these Fredholm operators is as operators that are "almost invertible". This notion of "almost invertible" will be made pre- cise later. For now, notice that that operator is almost injective as it has only a finite dimensional kernal, and almost surjective as kerT ∗ = (ImageT )? is also finite dimensional. We will now introduce some algebraic structure on the set of all bounded linear functionals, L(H) and from here make the above discussion precise. 2 2. Bounded Linear Operators as a Banach Algebra Definition 2.1:A C algebra is a ring A with identity along with a ring homomorphism f : C ! A such that 1 7! 1A and f(C) ⊆ Z(A). Observe that L(H) is a C algebra. We can add and multiply operators, and have additive inverses. We also have a zero element, the zero map, and an identity element, the identity map. Also, C sits inside of Z(L(H)) as the maps that are simply multiplication by a complex scalar. Definition 2.2: A Banach Algebra B is an algebra over C with identity which has a norm kk making it into a Banach space. Further we require that k1k = 1 and kfgk ≤ kfkkgk for all f; g 2 B. Indeed, L(H) is a Banach Algebra under the sup norm. We have seen that the sup norm satisfies the required properties. Definition 2.3: If U is a Banach Algebra, then an involution on U is a mapping T ! T ∗ such that: 1) T ∗∗ = T 2) (αS + βT )∗ =αS ¯ ∗ + βT¯ ∗ 3) (ST )∗ = T ∗S∗ The adjoint map on L(H) satisfies all of these properties as observed in class. Definition 2.4: A Banach Algebra with an involution map such that ∗ 2 ∗ kTT k = kT k is called a C − algebra. We notice that the adjoint map satisfies this final condition. Now we have our final algebraic structure on L(H). We can now say that L(H) is a ∗ C − algebra. Proposition 2.5: The set of all compact operators on H, denoted K(H), is a closed two-sided ideal in L(H). Proof:K(H) is closed as it is the closure of the finite rank operators. We have also seen that if S 2 L(H) and T 2 K(H) then ST 2 K(H) and TS 2 K(H). Furthermore, if T is a compact operator, then so is T ∗. 3 3. The Calkin Algebra Now that we have an algebra and a closed ideal it is very natural to consider the quotient algebra: L(H)=K(H). This quotient algebra is called the Calkin algebra, and it is again a (C)∗-algebra. We will prove that the invertible elements of the Calkin algebra are exactly the images of the Fred- holm operators under the quotient map. This result will show that Fredholm operators are invertible up to a compact operator exactly. First, we have to prove a few Lemmas. Lemma 3.1: The unit ball of a Hilbert space H,(H)1, is compact in norm topology if and only if H is finite dimensional. Proof: Suppose that H is finite dimensional. Then H is isometrically isomorphic to (C)n. Since the unit ball in (C)n is compact, the unit ball in H is also compact. Now assume that H is 1-dimensional. Then there exists an orthonormal 1 subset fengn=1p ⊂ (H)1. ken − emk = 2 for any n; m 2 N with n 6= m. Thus (H)1 cannot be compact. Lemma 3.2: If H is an 1-dimensional Hilbert space and T is a com- pact operator on H, then the range of T contains no closed 1-dimensional subspace. Proof: Let M be a closed subspace in the range of T . Let PM be the projection map onto M. Then PM T is compact. Define A : H ! M by Af = PM T f. Then A is bounded and onto. Thus by the open mapping theorem, A is an open map. Therefore, A((H)1) contains the open ball in M of radius δ for some small δ > 0 centered at 0. Now, since the closed ball of radius δ is contained in the compact set PM T ((H)1)) we get that M is finite dimensional by Lemma 3.2. Lemma 3.3: Let H be a Hilbert space, M a closed subspace of H, and 4 N a finite dimensional subspace of H. Then the linear space M + N is a closed subspace of H. Proof: Omitted. See Lemma 5.16 in 1. We will now state and prove the main result of this lecture. Theorem 3.4(Atkinson): If H is a Hilbert space, then T 2 L(H) is a Fredholm operator if and only if the image of T in the Calkin algebra is invertible. Proof: Suppose that the image of T is invertible in the Calkin algebra. Then there exists A 2 L(H) and K 2 K(H) such that AT = I + K. Let f 2 ker(I + K) then (I + K)f = 0. ) If = −Kf ) Kf = −f So, f 2 rangeK. Thus, kerT ⊂ kerAT = ker(I + K) ⊂ rangeK So, by Lemma 3.2 the dimension of kerT < 1 as rangeK contains no closed 1 dimensional subspace (kerT is closed as it is a kernal). Now, since T ∗ is also invertible in the Calkin algebra, dimkerT ∗ < 1 by symmetry. Next, we need to show that rangeT is closed in H. We know that the finite rank operators are dense in the compact operators, 1 so there exists F , a finite rank operator, such that kK − F k < 2 . Claim:T is bounded below on kerF . Let f 2 kerF . kAT fk = kf + Kfk = kf + Kf − F fk = kf − (Kf − F f)k 1 kfk ≥ kfk − kKf − F fk ≥ kfk − kK − F kkfk ≥ kfk − 2 kfk = 2 kfk Thus, 2 ≤ kAT fk ≤ kAkkT fk kfk ) 2kAk ≤ kT fk So, T is bounded below on kerF and we have proved the claim. Claim: T (kerF ) is a closed subspace of H. 1 Let fT fngn=1 be a Cauchy sequence in T (kerF ). Then kfn − fmk ≤ 2kAkkT fn − T fmk 5 ) ffng is Cauchy. So, if f = lim fn then T f = lim T fn is in T (kerF ). Therefore, T (kerF ) is closed to compelte the proof of the claim. Now, since F is a finite rank operator, we have that (ker(F )perp is finite dimensional. So, rangeT = T (kerF )+T ((kerF )perp) is a closed subspace of H by Lemma 3.3. Thus, T is a Fredholm operator. Now, assume that T is Fredholm operator. We will show that it is in- vertible up to a finite rank operator, and thus up to a compact operator. perp Define T0 :(kerT ) ! rangeT by T0f = T f. Claim: T0 is a bijection. First we will show that it is surjective. Let g 2 rangeT then there exists f 2 H such that T f = g. If f 2 (kerT )perp then we are done. perp If f2 = (kerT ) then T (f) = 0. In this case g = 0 and T (0) = 0 so T0 is surjective. perp Now we will show that it is injective. If T0f = T0g for f; g 2 (kerT ) . Then, T0(f − g) = 0 ) f − g = 0 so T0 is injective. Thus, T0 is bijective and so it has an inverse. −1 Now, define S : H ! H by Sf = T0 f if f 2 rangeT , and Sf = 0 if f 2 (rangeT )perp. −1 S is bounded since it is bounded on rangeT as it is equal to T0 there and zero otherwise. So, ST = I −P1 and TS = I −P2 where P1 is the projection map onto kerT perp ∗ and P2is the projection map onto (rangeT ) = kerT . Since dimkerT < 1P1 is a finite rank operator and thus is compact. Simi- ∗ larly as dimkerT < 1P2 is finite rank and thus compact. So, the image of S is the inverse of the image of T is the Calkin algebra. Therefore, T is invertible in the Calkin algebra as required. 6 There are a couple interesting things to note in this proof. First, we only used the fact that T had a one sided inverse in the first direction. Then, in the second direction we proved that T was invertible up to a finite rank operator instead of a compact operator. 4. Examples Example 4.1 If T is an invertible operator, then T is a Fredholm operator. This is clear, as if T is invertible then certainly the image of T in the Calkin algebra is invertible.